When a threshold is exceeded and the timer begins, at what rate is the CPU monitored/polled in order to see if it falls below the reset value?
It's polled with the same frequency. It's basically based on a watcher which can be an integer multiplier of the polling interval. So if you poll devices every 5 minutes it's going to clear the alarm in no less than 5 minutes.
Dit take the average value from the last 5 minutes, or just a instant value on the 5 minute mark? Does it know if the CPU dropped below 70% during the 5 minutes even if it jumps back up ?
It takes the instant value when it polls it (based on which OID is polled can make a slight difference, on some devices it might be the 1minAvg, 5minAvg and so on especially on Cisco if I remember correctly). So to make it clear, the value is read at the moment the device is polled but the read value can sometimes be an already average value. If it went below the threshold and above again in the 5 min interval won't be considered if the value read is still above the threshold at the moment it's polled.